November 24, 2017

Integrate Approximation Models Into the Workflow Easily

The release of pSeven 6.12 brings up a new block for working with approximation models — Approximation model. This block is intended to evaluate approximation models, for example, trained in pSeven or pSeven Core, and will replace the old ApproxPlayer block.

Key features of the new block:

  • Configurable Variables and Responses as interfaces to model inputs and outputs
  • Auto-configuration of Variables and Responses from model annotations
  • Variables and Responses with a defined name and size
  • Flexible run-time model export
  • Mappable model properties
  • Significant speedup of model evaluation in cycles

How to use?

Let’s consider that you have an approximation model trained in the Predictive Modeling Toolkit and exported to the file. For example, we will use one of the models trained in the “Build. Validate. Explore.” Tech Tips series (Part 1, Part 2, Part 3).

Load the model to the Approximation model block within the file dialog in the Configuration tab. After that, main information appears in the Model summary and Training options panes. The Model summary pane contains information about the technique, accuracy evaluation, smoothing and input and output size of the model. Training options pane shows training options of the model.

Let’s configure human-readable variables and responses. Add the following variables: “Nozzle Diameter”, “Nozzle Angle”, “Flow Velocity 1”, “Flow Velocity 2” with size 1 (default). Also, add responses with the names “Flow Temperature” and “Pressure Drop” and size 1.

Add the sample source, for example, the Design space exploration block with the same variables and responses. The bounds for “Nozzle Diameter”, “Flow Velocity 1” and “Flow Velocity 2” variables are: 0.05 for the lower bound, 0.2 for the upper bound. “Nozzle Angle”: 0.17 for the lower bound, 0.9 for the upper bound.

Create responses with default parameters. Link blocks with an autolink function. Uplink all ports with Design suffix in the Design of experiments block.

Let’s run the workflow and post-process the results. Add all outputs to the Report database. Select all data series in the Report database and create the Parallel coordinates chart to find the best design for your needs.

In Conclusion

The new block greatly contributed to more convenient integration of approximation models, thus they became an even more useful instrument for parametric studies and optimization runs in pSeven.

Interested in the solution?

Click to request a free 30-day demo.

Request demo